An exercise in high-quality DVD conversion

[diprotic]

I'm not in front of my encoding machine but that looks basically the same as mine as I recall. The only difference I'm sure of is that I set BITRATE to 8100 and MAXRATE to 9100 based on 70 minutes of video on DVD5 (audio at 384kbps) using this calculator: http://dvd-hq.info/bitrate_calculator.php. Seems strange that that would increase noise though. I'll try again at home and let you know how it goes.

[jagabo]

The only difference I'm sure of is that I set BITRATE to 8100 and MAXRATE to 9100

I don't think that's responsible for the difference. Part of the problem is that you decimated from 60 frames per second to 24 frames per second. Whereas I converted to 60 fields per second. Repeat your encoding with the new script and it should look better.

[diprotic]

You're right, there was something funky with my HC settings. I ran it again with the given .avs and it looks great, no noise and proper interlacing.

So, I've come a long way in my exercise in high-quality DVD encoding. I just have one more point I need clarification on:

Normally I work with FILM sources at 24000/1001 progressive. When converting to DVD, I just leave framerate intact and enable pulldown, easy.

This particular source is different: it's 59.94 progressive. From what jagabo says, I should not decimate to 23.976 progressive. Instead, according to poisondeathray, I can either drop it to 29.97 progressive using SelectEven(), or do what we just did, converting 59.94 progressive to 59.94 interlaced. I assume the hardware or software decoder playing the final DVD will decide how and when to deinterlace.

Final question: Is there a compelling reason to choose one method or the other, all things considered?

I grew up thinking "progressive is better than interlaced" but I can't back that up empirically. What do the experts say?

[poisondeathray]

Your first source was 59.94p but it had 2:3 repeats. e.g. aaabbcccddeeeff... So in fact it was 23.976 content . So you should decimate to 23.976 so there are no repeats (every frame is a unique frame)

This last example has 59.94 unique frames per second (no repeats). So it is a 59.94p source. So ideally you want to keep 59.94 progressive frames per second. But - It's not possible in DVD, so you have to either a) convert to interlaced to keep the motion, or b) drop 1/2 the motion and make a 29.97p DVD

Progressive is better than interlaced ; But there are a lot of "BUTS" to that blanketed statemnt - ONLY if you have sufficient bandwidth and the same temporal resolution. ie. 480p59.94 with adequate bitate will always be better than 480i59.94 (again this is also called 480i29.97, same thing, differernt naming conventions)

Interlaced in NTSC regions is 59.94 unique fields per second. Each field is like 1/2 a frame. Only 1/2 the spatial resolution , also, when viewing on progressive (flat panel) displays, it will be deinterlaced - this predisposes you to deinterlacing artifacts (jaggies, aliasing, "marching ants") . The quality of the deinterlacing varies, some HDTV's are fairly good, some are not good.

With DVD and a 59.94p source, essentially you're trading off 1/2 the temporal resolution or 1/2 the spatial resolution; there is no "best" . For things like sports, it's a no brainer - you usually want to preserve the motion and so you make an interlaced DVD. 29.97p will look way too choppy for that type of content. I would say most of the time people will make interlaced DVD from a 59.94p source.

I assume the hardware or software decoder playing the final DVD will decide how and when to deinterlace.

Hardware path (Standalone DVD or blu-ray player, TV setup) does for a properly authored DVD; but for software playback on a computer with a physical DVD disc (or mounted ISO) it depends - if you use DVD playing software (e.g. PowerDVD, WinDVD), then yes. They will even IVTC 24p soft telecined DVD's. But many other non DVD playing software (e.g. software media players) won't automatically do the proper processing (e.g. they might deinterlace when they shouldn't, or only single rate deinterlace instead of "bobbing")

[diprotic]

Your first source was 59.94p but it had 2:3 repeats. e.g. aaabbcccddeeeff... So in fact it was 23.976 content . So you should decimate to 23.976 so there are no repeats (every frame is a unique frame)

This last example has 59.94 unique frames per second (no repeats). So it is a 59.94p source. So ideally you want to keep 59.94 progressive frames per second. But - It's not possible in DVD, so you have to either a) convert to interlaced to keep the motion, or b) drop 1/2 the motion and make a 29.97p DVD

Progressive is better than interlaced ; But there are a lot of "BUTS" to that blanketed statemnt - ONLY if you have sufficient bandwidth and the same temporal resolution. ie. 480p59.94 with adequate bitate will always be better than 480i59.94 (again this is also called 480i29.97, same thing, diffeernt naming conventions)

Interlaced in NTSC regions is 59.94 unique fields per second. Each field is like 1/2 a frame. Only 1/2 the spatial resolution , also, when viewing on progressive (flat panel) displays, it will be deinterlaced - this predisposes you to deinterlacing artifacts (jaggies, aliasing, "marching ants") . The quality of the deinterlacing varies, some HDTV's are fairly good, some are not good.

With DVD and a 59.94p source, essentially you're trading off 1/2 the temporal resolution or 1/2 the spatial resolution; there is no "best" . For things like sports, it's a no brainer - you usually want to preserve the motion and so you make an interlaced DVD. I would say most of the time people will make interlaced DVD from a 59.94p source.

I'm tracking with you. You're saying I just have a decision to make: what's more important: spatial or temporal resolution? In sports and other high-motion content, temporal is more important. But in my case, since this is just a speaker up on stage, I think spatial resolution is more important.

So if I value spatial resolution in this case, with this particular source and destination, I should go the 29.97p route. Is this a good avs?

ffVideoSource("sample.mp4", fpsden=1001, fpsnum=60000)
BilinearResize(720,480)
ColorMatrix(mode="Rec.601->Rec.709")
SelectEven()

[poisondeathray]

I'm tracking with you. You're saying I just have a decision to make: what's more important: spatial or temporal resolution? In sports and other high-motion content, temporal is more important. But in my case, since this is just a speaker up on stage, I think spatial resolution is more important.

So if I value spatial resolution in this case, with this particular source and destination, I should go the 29.97p route. Is this a good avs?

ffVideoSource("sample.mp4", fpsden=1001, fpsnum=60000)
BilinearResize(720,480)
ColorMatrix(mode="Rec.601->Rec.709")
SelectEven()

That would work for a "29.97p" DVD

And yes, for this type of talking /interview content, motion is less important than for something like sports. (But some speakers are very dynamic, flapping their wings and strutting on stage - something to keep in mind)

This is a useful learning experience for you - do some quick encodes either way and see what you prefer . Make sure you view it properly (e.g. if you are using a software media player, it must be set to "bob" deinterlace or double rate deinterlace to emulate what you would see on a normal DVD setup in terms of motion for the interlaced version)

(And I would always add clamp=0 for colormatrix)

[diprotic]

This is a useful learning experience for you - do some quick encodes either way and see what you prefer . Make sure you view it properly (e.g. if you are using a software media player, it must be set to "bob" deinterlace or double rate deinterlace to emulate what you would see on a normal DVD setup in terms of motion for the interlaced version)

I quite enjoy it I'm previewing with VLC. Is yadif(x2) a double rate deinterlacer?

(And I would always add clamp=0 for colormatrix)

Oh right, I forgot, thanks.

So here are my two options:

29.97p:

ffVideoSource("sample.mp4", fpsden=1001, fpsnum=60000)
BilinearResize(720,480)
ColorMatrix(mode="Rec.601->Rec.709", clamp=0)
SelectEven()

59.94i/29.97i:

ffVideoSource("sample.mp4", fpsden=1001, fpsnum=60000)
BilinearResize(720,480)
ColorMatrix(mode="Rec.601->Rec.709", clamp=0)
AssumeTFF()
SeparateFields()
SelectEvery(4,0,3)
Weave()

Look good?

[poisondeathray]

Yes looks good. Yadif(2x) in VLC is bobbing. For viewing the 29.97p encode in VLC, don't forget to disable deinterlacing .

"Bob" is probably closer to what most HDTV's use . Some more expensive sets add other processing and have better deinterlacing


BTW, You can also preview avs scripts in a media player directly (most PC software media players will open avs scripts)

So you can add Bob() or Yadif(1,1) to the bottom of the re-interlaced script and play that in a media player to preview if you like. The 29.97p script you can play as is

[diprotic]

Sweet, even better. I'm curious to do some a/b testing and see which I like best, though I'm almost certain progressive is best here.

For the final mux, any special instructions?

The way I'm used to doing it is:

ffmpeg -i video.m2v -i audio.ac3 -map 0:0 -map 1:0 -c copy final.mpg

[poisondeathray]

I wouldn't use ffmpeg for the final multiplex (for DVD authoring) ; unless something has changed it doesn't do strict DVD compliant muxes

If it's just to play the file, then that will work

[diprotic]

I wouldn't use ffmpeg for the final multiplex (for DVD authoring) ; unless something has changed it doesn't do strict DVD compliant muxes

If it's just to play the file, then that will work

How then do you suggest I mux for sure DVD compliance?

[poisondeathray]

I wouldn't use ffmpeg for the final multiplex (for DVD authoring) ; unless something has changed it doesn't do strict DVD compliant muxes

If it's just to play the file, then that will work

How then do you suggest I mux for sure DVD compliance?

Often playback problems and glitches on a DVD can be traced back to a poor muxing job

I would use a dedicated authoring tool, or muxman if you just need a muxer (in fact, many authoring GUI's use muxman) . Or retail ones like scenarist, encore, dvda, dvdlab pro etc...

https://www.videohelp.com/tools/sections/authoring-dvd

[diprotic]

I prefer open source tools because I run linux/mac a lot of the time, but I'm running HC in a VM so it looks like muxman may be the ticket. I'm just interested in the simplest way to get a proper mux.

[jagabo]

You're saying I just have a decision to make: what's more important: spatial or temporal resolution?

Take a look at 24v30v60.avi in this post:

https://forum.videohelp.com/threads/307004-Best-framerate-conversion-%28eg-23-97-to-30-...=1#post1888926

View it full screen for maximum effect.

[diprotic]

I tried this avs file:

ffVideoSource("sample.mp4", fpsden=1001, fpsnum=60000) BilinearResize(720,480) ColorMatrix(mode="Rec.601->Rec.709", clamp=0) SelectEven()

and got this result from HC:

error loading Avisynth script,Avisynth error: the named argument "clamp" to ColorMatrix had the wrong type.

Any ideas?

[poisondeathray]

error loading Avisynth script,Avisynth error: the named argument "clamp" to ColorMatrix had the wrong type.

you're probably using an old colormatrix version

v2.5
http://web.missouri.edu/~kes25c/
http://web.missouri.edu/~kes25c/ColorMatrixv25.zip


The other DVD muxer that linux users frequently use is mplex; but it's reported to have issues with certain DVD players . I would stick with muxman
http://linux.die.net/man/1/mplex

[diprotic]

I've finished my project and want to share the sum of my experience, for my own reference and for anyone interested.

I started with an exercise in high-quality DVD conversion. What I forgot to mention is that I'm confining myself to free tools, since I'm sure I could achieve even higher quality with expensive hardware and software.

For my purposes, I prefer "free" to mean "free and open source software," but I found out that, despite many allegedly mature and active projects, in many cases they don't measure up to other free closed-source programs.

My end-result was a fully-authored DVD-5 with a custom menu and 2 titles—one transcoded, the other simply extracted from an existing DVD ISO.

Here are the (free) tools and techniques I used:

• Video Transcoder: HCEnc — I started with dogged determination to use either ffmpeg or mencoder. I tried many, many, many incantations (review this thread) to try and achieve a high-quality, DVD-compliant conversion. In the end, both the resulting bitrate and framerate left much to be desired. HCEnc paired with Avisynth left nothing to be desired—it fully used the allocated bitrate and resulted in a proper framerate. From now on, all my MPEG2 is going through HC.
• Demultiplexer: ffmpeg — As far as I can tell, ffmpeg does a fine job of copying out a video or audio stream from a program stream.
• Audio Normalizer: Audacity — In this project, both titles' audio needed normalization. I extracted the audio to raw PCM/WAV with ffmpeg, then normalized in Audacity, saving back to PCM/WAV.
• Audio Transcoder: Aften — Once I had the final normalized WAV, I re-compressed it to AC3 with Aften.
• Bitrate Calculator: http://dvd-hq.info/bitrate_calculator.php — It made the job easy and it definitely worked. The resulting files were exactly the right size for DVD-5, with just enough headroom for menus, authoring, etc.
• Extract MPEG from DVD ISO: mplayer — mplayer -dumpstream -dumpfile output.vob -dvd-device source.iso dvd://1
• Multiplexer: MuxMan — mplex says it right in the help output: their DVD profile isn't mature yet. MuxMan is free and apparently quite DVD-compliant. My only frustration is that it automatically splits the resulting VOB files into 1024MB chunks, which in my case wasn't what I wanted right away. So, I had the need for:
• VOB joiner: cat — Humble GNU cat joined my VOBs just fine.
• DVD Authoring: DVDStyler — Certainly not the best tool available for all-around authoring, but really I only used it for creating a custom menu and wrapping the whole thing up in an ISO. I told it not to touch the 2 titles which had already been properly transcoded and muxed. If there ends up being anything not strictly DVD-compliant in my project, it probably has to do with DVDStyler, but there really aren't any good free cross-platform authoring programs. Ultimately I'm very satisfied with the result, and I've created several projects with DVDStyler that have had no problem playing in standalone players. For now, this is as good as it gets at a $0 price tag.

I'm sure I've messed something up, as video seems to be an exercise in masochism...but for now, I'm very happy I learned a lot, and I hope to help others do the same!


Thanks all!

[manono]

If there ends up being anything not strictly DVD-compliant in my project, it probably has to do with DVDStyler, but there really aren't any good free cross-platform authoring programs.

I don't think DVD Styler uses Muxman as its authoring program, and Muxman is as good as any out there, free or commercial. The one it does use (spumux) isn't nearly as good, but I've never had any problems in playing the resulting DVDs. However, using DVD Styler's sister menu-making and authoring program, GUI4DVDAuthor, allows you to both make menus and choose to use Muxman as the authoring engine.

Nice description of how you accomplished your project.

[jagabo]

Yes, nice mini guide. One thing I might add since you use Audacity to normalize your audio: Audacity with the ffmpeg plugin can import audio directly from most media files. And it can export as AC3. That could eliminate a couple of your steps.

[diprotic]

If there ends up being anything not strictly DVD-compliant in my project, it probably has to do with DVDStyler, but there really aren't any good free cross-platform authoring programs.

I don't think DVD Styler uses Muxman as its authoring program, and Muxman is as good as any out there, free or commercial. The one it does use (spumux) isn't nearly as good, but I've never had any problems in playing the resulting DVDs. However, using DVD Styler's sister menu-making and authoring program, GUI4DVDAuthor, allows you to both make menus and choose to use Muxman as the authoring engine.

Nice description of how you accomplished your project.

You're right—DVDStyler doesn't use MuxMan. I already muxed the .ac3 and .m2v with MuxMan, then joined the resulting VOBs with cat. I'm sure it's the long way around. Then I told DVDStyler to leave the source files untouched.

Interesting about GUI4DVDAuthor, I'll need to look into that. However it does appear to be Windows only, which in my case is a minus. Thanks though!